2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
92 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
115 #include <asm/uaccess.h>
116 #include <asm/system.h>
118 #include <linux/netdevice.h>
119 #include <net/protocol.h>
120 #include <linux/skbuff.h>
121 #include <net/net_namespace.h>
122 #include <net/request_sock.h>
123 #include <net/sock.h>
124 #include <linux/net_tstamp.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
127 #include <net/cls_cgroup.h>
129 #include <linux/filter.h>
131 #include <trace/events/sock.h>
138 * Each address family might have different locking rules, so we have
139 * one slock key per address family:
141 static struct lock_class_key af_family_keys[AF_MAX];
142 static struct lock_class_key af_family_slock_keys[AF_MAX];
145 * Make lock validator output more readable. (we pre-construct these
146 * strings build-time, so that runtime initialization of socket
149 static const char *const af_family_key_strings[AF_MAX+1] = {
150 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
151 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
152 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
153 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
154 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
155 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
156 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
157 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
158 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
159 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
160 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
161 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
162 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
163 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
165 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
166 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
167 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
168 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
169 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
170 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
171 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
172 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
173 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
174 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
175 "slock-27" , "slock-28" , "slock-AF_CAN" ,
176 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
177 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
178 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
179 "slock-AF_NFC" , "slock-AF_MAX"
181 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
182 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
183 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
184 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
185 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
186 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
187 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
188 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
189 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
190 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
191 "clock-27" , "clock-28" , "clock-AF_CAN" ,
192 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
193 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
194 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
195 "clock-AF_NFC" , "clock-AF_MAX"
199 * sk_callback_lock locking rules are per-address-family,
200 * so split the lock classes by using a per-AF key:
202 static struct lock_class_key af_callback_keys[AF_MAX];
204 /* Take into consideration the size of the struct sk_buff overhead in the
205 * determination of these values, since that is non-constant across
206 * platforms. This makes socket queueing behavior and performance
207 * not depend upon such differences.
209 #define _SK_MEM_PACKETS 256
210 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
211 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
212 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
214 /* Run time adjustable parameters. */
215 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
216 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
217 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
218 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
220 /* Maximal space eaten by iovec or ancillary data plus some space */
221 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
222 EXPORT_SYMBOL(sysctl_optmem_max);
224 #if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
225 int net_cls_subsys_id = -1;
226 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
229 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
233 if (optlen < sizeof(tv))
235 if (copy_from_user(&tv, optval, sizeof(tv)))
237 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
241 static int warned __read_mostly;
244 if (warned < 10 && net_ratelimit()) {
246 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
247 "tries to set negative timeout\n",
248 current->comm, task_pid_nr(current));
252 *timeo_p = MAX_SCHEDULE_TIMEOUT;
253 if (tv.tv_sec == 0 && tv.tv_usec == 0)
255 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
256 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
260 static void sock_warn_obsolete_bsdism(const char *name)
263 static char warncomm[TASK_COMM_LEN];
264 if (strcmp(warncomm, current->comm) && warned < 5) {
265 strcpy(warncomm, current->comm);
266 printk(KERN_WARNING "process `%s' is using obsolete "
267 "%s SO_BSDCOMPAT\n", warncomm, name);
272 static void sock_disable_timestamp(struct sock *sk, int flag)
274 if (sock_flag(sk, flag)) {
275 sock_reset_flag(sk, flag);
276 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
277 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
278 net_disable_timestamp();
284 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
289 struct sk_buff_head *list = &sk->sk_receive_queue;
291 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
292 number of warnings when compiling with -W --ANK
294 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
295 (unsigned)sk->sk_rcvbuf) {
296 atomic_inc(&sk->sk_drops);
297 trace_sock_rcvqueue_full(sk, skb);
301 err = sk_filter(sk, skb);
305 if (!sk_rmem_schedule(sk, skb->truesize)) {
306 atomic_inc(&sk->sk_drops);
311 skb_set_owner_r(skb, sk);
313 /* Cache the SKB length before we tack it onto the receive
314 * queue. Once it is added it no longer belongs to us and
315 * may be freed by other threads of control pulling packets
320 /* we escape from rcu protected region, make sure we dont leak
325 spin_lock_irqsave(&list->lock, flags);
326 skb->dropcount = atomic_read(&sk->sk_drops);
327 __skb_queue_tail(list, skb);
328 spin_unlock_irqrestore(&list->lock, flags);
330 if (!sock_flag(sk, SOCK_DEAD))
331 sk->sk_data_ready(sk, skb_len);
334 EXPORT_SYMBOL(sock_queue_rcv_skb);
336 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
338 int rc = NET_RX_SUCCESS;
340 if (sk_filter(sk, skb))
341 goto discard_and_relse;
345 if (sk_rcvqueues_full(sk, skb)) {
346 atomic_inc(&sk->sk_drops);
347 goto discard_and_relse;
350 bh_lock_sock_nested(sk);
353 if (!sock_owned_by_user(sk)) {
355 * trylock + unlock semantics:
357 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
359 rc = sk_backlog_rcv(sk, skb);
361 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
362 } else if (sk_add_backlog(sk, skb)) {
364 atomic_inc(&sk->sk_drops);
365 goto discard_and_relse;
376 EXPORT_SYMBOL(sk_receive_skb);
378 void sk_reset_txq(struct sock *sk)
380 sk_tx_queue_clear(sk);
382 EXPORT_SYMBOL(sk_reset_txq);
384 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
386 struct dst_entry *dst = __sk_dst_get(sk);
388 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
389 sk_tx_queue_clear(sk);
390 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
397 EXPORT_SYMBOL(__sk_dst_check);
399 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
401 struct dst_entry *dst = sk_dst_get(sk);
403 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
411 EXPORT_SYMBOL(sk_dst_check);
413 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
415 int ret = -ENOPROTOOPT;
416 #ifdef CONFIG_NETDEVICES
417 struct net *net = sock_net(sk);
418 char devname[IFNAMSIZ];
423 if (!capable(CAP_NET_RAW))
430 /* Bind this socket to a particular device like "eth0",
431 * as specified in the passed interface name. If the
432 * name is "" or the option length is zero the socket
435 if (optlen > IFNAMSIZ - 1)
436 optlen = IFNAMSIZ - 1;
437 memset(devname, 0, sizeof(devname));
440 if (copy_from_user(devname, optval, optlen))
444 if (devname[0] != '\0') {
445 struct net_device *dev;
448 dev = dev_get_by_name_rcu(net, devname);
450 index = dev->ifindex;
458 sk->sk_bound_dev_if = index;
470 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
473 sock_set_flag(sk, bit);
475 sock_reset_flag(sk, bit);
479 * This is meant for all protocols to use and covers goings on
480 * at the socket level. Everything here is generic.
483 int sock_setsockopt(struct socket *sock, int level, int optname,
484 char __user *optval, unsigned int optlen)
486 struct sock *sk = sock->sk;
493 * Options without arguments
496 if (optname == SO_BINDTODEVICE)
497 return sock_bindtodevice(sk, optval, optlen);
499 if (optlen < sizeof(int))
502 if (get_user(val, (int __user *)optval))
505 valbool = val ? 1 : 0;
511 if (val && !capable(CAP_NET_ADMIN))
514 sock_valbool_flag(sk, SOCK_DBG, valbool);
517 sk->sk_reuse = valbool;
526 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
529 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
532 /* Don't error on this BSD doesn't and if you think
533 about it this is right. Otherwise apps have to
534 play 'guess the biggest size' games. RCVBUF/SNDBUF
535 are treated in BSD as hints */
537 if (val > sysctl_wmem_max)
538 val = sysctl_wmem_max;
540 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
541 if ((val * 2) < SOCK_MIN_SNDBUF)
542 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
544 sk->sk_sndbuf = val * 2;
547 * Wake up sending tasks if we
550 sk->sk_write_space(sk);
554 if (!capable(CAP_NET_ADMIN)) {
561 /* Don't error on this BSD doesn't and if you think
562 about it this is right. Otherwise apps have to
563 play 'guess the biggest size' games. RCVBUF/SNDBUF
564 are treated in BSD as hints */
566 if (val > sysctl_rmem_max)
567 val = sysctl_rmem_max;
569 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
571 * We double it on the way in to account for
572 * "struct sk_buff" etc. overhead. Applications
573 * assume that the SO_RCVBUF setting they make will
574 * allow that much actual data to be received on that
577 * Applications are unaware that "struct sk_buff" and
578 * other overheads allocate from the receive buffer
579 * during socket buffer allocation.
581 * And after considering the possible alternatives,
582 * returning the value we actually used in getsockopt
583 * is the most desirable behavior.
585 if ((val * 2) < SOCK_MIN_RCVBUF)
586 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
588 sk->sk_rcvbuf = val * 2;
592 if (!capable(CAP_NET_ADMIN)) {
600 if (sk->sk_protocol == IPPROTO_TCP)
601 tcp_set_keepalive(sk, valbool);
603 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
607 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
611 sk->sk_no_check = valbool;
615 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
616 sk->sk_priority = val;
622 if (optlen < sizeof(ling)) {
623 ret = -EINVAL; /* 1003.1g */
626 if (copy_from_user(&ling, optval, sizeof(ling))) {
631 sock_reset_flag(sk, SOCK_LINGER);
633 #if (BITS_PER_LONG == 32)
634 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
635 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
638 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
639 sock_set_flag(sk, SOCK_LINGER);
644 sock_warn_obsolete_bsdism("setsockopt");
649 set_bit(SOCK_PASSCRED, &sock->flags);
651 clear_bit(SOCK_PASSCRED, &sock->flags);
657 if (optname == SO_TIMESTAMP)
658 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
660 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
661 sock_set_flag(sk, SOCK_RCVTSTAMP);
662 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
664 sock_reset_flag(sk, SOCK_RCVTSTAMP);
665 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
669 case SO_TIMESTAMPING:
670 if (val & ~SOF_TIMESTAMPING_MASK) {
674 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
675 val & SOF_TIMESTAMPING_TX_HARDWARE);
676 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
677 val & SOF_TIMESTAMPING_TX_SOFTWARE);
678 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
679 val & SOF_TIMESTAMPING_RX_HARDWARE);
680 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
681 sock_enable_timestamp(sk,
682 SOCK_TIMESTAMPING_RX_SOFTWARE);
684 sock_disable_timestamp(sk,
685 SOCK_TIMESTAMPING_RX_SOFTWARE);
686 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
687 val & SOF_TIMESTAMPING_SOFTWARE);
688 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
689 val & SOF_TIMESTAMPING_SYS_HARDWARE);
690 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
691 val & SOF_TIMESTAMPING_RAW_HARDWARE);
697 sk->sk_rcvlowat = val ? : 1;
701 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
705 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
708 case SO_ATTACH_FILTER:
710 if (optlen == sizeof(struct sock_fprog)) {
711 struct sock_fprog fprog;
714 if (copy_from_user(&fprog, optval, sizeof(fprog)))
717 ret = sk_attach_filter(&fprog, sk);
721 case SO_DETACH_FILTER:
722 ret = sk_detach_filter(sk);
727 set_bit(SOCK_PASSSEC, &sock->flags);
729 clear_bit(SOCK_PASSSEC, &sock->flags);
732 if (!capable(CAP_NET_ADMIN))
738 /* We implement the SO_SNDLOWAT etc to
739 not be settable (1003.1g 5.3) */
741 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
745 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
755 EXPORT_SYMBOL(sock_setsockopt);
758 void cred_to_ucred(struct pid *pid, const struct cred *cred,
761 ucred->pid = pid_vnr(pid);
762 ucred->uid = ucred->gid = -1;
764 struct user_namespace *current_ns = current_user_ns();
766 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
767 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
770 EXPORT_SYMBOL_GPL(cred_to_ucred);
772 int sock_getsockopt(struct socket *sock, int level, int optname,
773 char __user *optval, int __user *optlen)
775 struct sock *sk = sock->sk;
783 int lv = sizeof(int);
786 if (get_user(len, optlen))
791 memset(&v, 0, sizeof(v));
795 v.val = sock_flag(sk, SOCK_DBG);
799 v.val = sock_flag(sk, SOCK_LOCALROUTE);
803 v.val = !!sock_flag(sk, SOCK_BROADCAST);
807 v.val = sk->sk_sndbuf;
811 v.val = sk->sk_rcvbuf;
815 v.val = sk->sk_reuse;
819 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
827 v.val = sk->sk_protocol;
831 v.val = sk->sk_family;
835 v.val = -sock_error(sk);
837 v.val = xchg(&sk->sk_err_soft, 0);
841 v.val = !!sock_flag(sk, SOCK_URGINLINE);
845 v.val = sk->sk_no_check;
849 v.val = sk->sk_priority;
854 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
855 v.ling.l_linger = sk->sk_lingertime / HZ;
859 sock_warn_obsolete_bsdism("getsockopt");
863 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
864 !sock_flag(sk, SOCK_RCVTSTAMPNS);
868 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
871 case SO_TIMESTAMPING:
873 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
874 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
875 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
876 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
877 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
878 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
879 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
880 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
881 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
882 v.val |= SOF_TIMESTAMPING_SOFTWARE;
883 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
884 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
885 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
886 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
890 lv = sizeof(struct timeval);
891 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
895 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
896 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
901 lv = sizeof(struct timeval);
902 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
906 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
907 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
912 v.val = sk->sk_rcvlowat;
920 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
925 struct ucred peercred;
926 if (len > sizeof(peercred))
927 len = sizeof(peercred);
928 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
929 if (copy_to_user(optval, &peercred, len))
938 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
942 if (copy_to_user(optval, address, len))
947 /* Dubious BSD thing... Probably nobody even uses it, but
948 * the UNIX standard wants it for whatever reason... -DaveM
951 v.val = sk->sk_state == TCP_LISTEN;
955 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
959 return security_socket_getpeersec_stream(sock, optval, optlen, len);
966 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
970 v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
979 if (copy_to_user(optval, &v, len))
982 if (put_user(len, optlen))
988 * Initialize an sk_lock.
990 * (We also register the sk_lock with the lock validator.)
992 static inline void sock_lock_init(struct sock *sk)
994 sock_lock_init_class_and_name(sk,
995 af_family_slock_key_strings[sk->sk_family],
996 af_family_slock_keys + sk->sk_family,
997 af_family_key_strings[sk->sk_family],
998 af_family_keys + sk->sk_family);
1002 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1003 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1004 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1006 static void sock_copy(struct sock *nsk, const struct sock *osk)
1008 #ifdef CONFIG_SECURITY_NETWORK
1009 void *sptr = nsk->sk_security;
1011 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1013 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1014 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1016 #ifdef CONFIG_SECURITY_NETWORK
1017 nsk->sk_security = sptr;
1018 security_sk_clone(osk, nsk);
1023 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1024 * un-modified. Special care is taken when initializing object to zero.
1026 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1028 if (offsetof(struct sock, sk_node.next) != 0)
1029 memset(sk, 0, offsetof(struct sock, sk_node.next));
1030 memset(&sk->sk_node.pprev, 0,
1031 size - offsetof(struct sock, sk_node.pprev));
1034 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1036 unsigned long nulls1, nulls2;
1038 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1039 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1040 if (nulls1 > nulls2)
1041 swap(nulls1, nulls2);
1044 memset((char *)sk, 0, nulls1);
1045 memset((char *)sk + nulls1 + sizeof(void *), 0,
1046 nulls2 - nulls1 - sizeof(void *));
1047 memset((char *)sk + nulls2 + sizeof(void *), 0,
1048 size - nulls2 - sizeof(void *));
1050 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1052 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1056 struct kmem_cache *slab;
1060 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1063 if (priority & __GFP_ZERO) {
1065 prot->clear_sk(sk, prot->obj_size);
1067 sk_prot_clear_nulls(sk, prot->obj_size);
1070 sk = kmalloc(prot->obj_size, priority);
1073 kmemcheck_annotate_bitfield(sk, flags);
1075 if (security_sk_alloc(sk, family, priority))
1078 if (!try_module_get(prot->owner))
1080 sk_tx_queue_clear(sk);
1086 security_sk_free(sk);
1089 kmem_cache_free(slab, sk);
1095 static void sk_prot_free(struct proto *prot, struct sock *sk)
1097 struct kmem_cache *slab;
1098 struct module *owner;
1100 owner = prot->owner;
1103 security_sk_free(sk);
1105 kmem_cache_free(slab, sk);
1111 #ifdef CONFIG_CGROUPS
1112 void sock_update_classid(struct sock *sk)
1116 rcu_read_lock(); /* doing current task, which cannot vanish. */
1117 classid = task_cls_classid(current);
1119 if (classid && classid != sk->sk_classid)
1120 sk->sk_classid = classid;
1122 EXPORT_SYMBOL(sock_update_classid);
1126 * sk_alloc - All socket objects are allocated here
1127 * @net: the applicable net namespace
1128 * @family: protocol family
1129 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1130 * @prot: struct proto associated with this new sock instance
1132 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1137 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1139 sk->sk_family = family;
1141 * See comment in struct sock definition to understand
1142 * why we need sk_prot_creator -acme
1144 sk->sk_prot = sk->sk_prot_creator = prot;
1146 sock_net_set(sk, get_net(net));
1147 atomic_set(&sk->sk_wmem_alloc, 1);
1149 sock_update_classid(sk);
1154 EXPORT_SYMBOL(sk_alloc);
1156 static void __sk_free(struct sock *sk)
1158 struct sk_filter *filter;
1160 if (sk->sk_destruct)
1161 sk->sk_destruct(sk);
1163 filter = rcu_dereference_check(sk->sk_filter,
1164 atomic_read(&sk->sk_wmem_alloc) == 0);
1166 sk_filter_uncharge(sk, filter);
1167 RCU_INIT_POINTER(sk->sk_filter, NULL);
1170 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1171 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1173 if (atomic_read(&sk->sk_omem_alloc))
1174 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1175 __func__, atomic_read(&sk->sk_omem_alloc));
1177 if (sk->sk_peer_cred)
1178 put_cred(sk->sk_peer_cred);
1179 put_pid(sk->sk_peer_pid);
1180 put_net(sock_net(sk));
1181 sk_prot_free(sk->sk_prot_creator, sk);
1184 void sk_free(struct sock *sk)
1187 * We subtract one from sk_wmem_alloc and can know if
1188 * some packets are still in some tx queue.
1189 * If not null, sock_wfree() will call __sk_free(sk) later
1191 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1194 EXPORT_SYMBOL(sk_free);
1197 * Last sock_put should drop reference to sk->sk_net. It has already
1198 * been dropped in sk_change_net. Taking reference to stopping namespace
1200 * Take reference to a socket to remove it from hash _alive_ and after that
1201 * destroy it in the context of init_net.
1203 void sk_release_kernel(struct sock *sk)
1205 if (sk == NULL || sk->sk_socket == NULL)
1209 sock_release(sk->sk_socket);
1210 release_net(sock_net(sk));
1211 sock_net_set(sk, get_net(&init_net));
1214 EXPORT_SYMBOL(sk_release_kernel);
1217 * sk_clone_lock - clone a socket, and lock its clone
1218 * @sk: the socket to clone
1219 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1221 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1223 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1227 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1228 if (newsk != NULL) {
1229 struct sk_filter *filter;
1231 sock_copy(newsk, sk);
1234 get_net(sock_net(newsk));
1235 sk_node_init(&newsk->sk_node);
1236 sock_lock_init(newsk);
1237 bh_lock_sock(newsk);
1238 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1239 newsk->sk_backlog.len = 0;
1241 atomic_set(&newsk->sk_rmem_alloc, 0);
1243 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1245 atomic_set(&newsk->sk_wmem_alloc, 1);
1246 atomic_set(&newsk->sk_omem_alloc, 0);
1247 skb_queue_head_init(&newsk->sk_receive_queue);
1248 skb_queue_head_init(&newsk->sk_write_queue);
1249 #ifdef CONFIG_NET_DMA
1250 skb_queue_head_init(&newsk->sk_async_wait_queue);
1253 spin_lock_init(&newsk->sk_dst_lock);
1254 rwlock_init(&newsk->sk_callback_lock);
1255 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1256 af_callback_keys + newsk->sk_family,
1257 af_family_clock_key_strings[newsk->sk_family]);
1259 newsk->sk_dst_cache = NULL;
1260 newsk->sk_wmem_queued = 0;
1261 newsk->sk_forward_alloc = 0;
1262 newsk->sk_send_head = NULL;
1263 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1265 sock_reset_flag(newsk, SOCK_DONE);
1266 skb_queue_head_init(&newsk->sk_error_queue);
1268 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1270 sk_filter_charge(newsk, filter);
1272 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1273 /* It is still raw copy of parent, so invalidate
1274 * destructor and make plain sk_free() */
1275 newsk->sk_destruct = NULL;
1276 bh_unlock_sock(newsk);
1283 newsk->sk_priority = 0;
1285 * Before updating sk_refcnt, we must commit prior changes to memory
1286 * (Documentation/RCU/rculist_nulls.txt for details)
1289 atomic_set(&newsk->sk_refcnt, 2);
1292 * Increment the counter in the same struct proto as the master
1293 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1294 * is the same as sk->sk_prot->socks, as this field was copied
1297 * This _changes_ the previous behaviour, where
1298 * tcp_create_openreq_child always was incrementing the
1299 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1300 * to be taken into account in all callers. -acme
1302 sk_refcnt_debug_inc(newsk);
1303 sk_set_socket(newsk, NULL);
1304 newsk->sk_wq = NULL;
1306 if (newsk->sk_prot->sockets_allocated)
1307 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1309 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1310 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1311 net_enable_timestamp();
1316 EXPORT_SYMBOL_GPL(sk_clone_lock);
1318 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1320 __sk_dst_set(sk, dst);
1321 sk->sk_route_caps = dst->dev->features;
1322 if (sk->sk_route_caps & NETIF_F_GSO)
1323 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1324 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1325 if (sk_can_gso(sk)) {
1326 if (dst->header_len) {
1327 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1329 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1330 sk->sk_gso_max_size = dst->dev->gso_max_size;
1334 EXPORT_SYMBOL_GPL(sk_setup_caps);
1336 void __init sk_init(void)
1338 if (totalram_pages <= 4096) {
1339 sysctl_wmem_max = 32767;
1340 sysctl_rmem_max = 32767;
1341 sysctl_wmem_default = 32767;
1342 sysctl_rmem_default = 32767;
1343 } else if (totalram_pages >= 131072) {
1344 sysctl_wmem_max = 131071;
1345 sysctl_rmem_max = 131071;
1350 * Simple resource managers for sockets.
1355 * Write buffer destructor automatically called from kfree_skb.
1357 void sock_wfree(struct sk_buff *skb)
1359 struct sock *sk = skb->sk;
1360 unsigned int len = skb->truesize;
1362 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1364 * Keep a reference on sk_wmem_alloc, this will be released
1365 * after sk_write_space() call
1367 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1368 sk->sk_write_space(sk);
1372 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1373 * could not do because of in-flight packets
1375 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1378 EXPORT_SYMBOL(sock_wfree);
1381 * Read buffer destructor automatically called from kfree_skb.
1383 void sock_rfree(struct sk_buff *skb)
1385 struct sock *sk = skb->sk;
1386 unsigned int len = skb->truesize;
1388 atomic_sub(len, &sk->sk_rmem_alloc);
1389 sk_mem_uncharge(sk, len);
1391 EXPORT_SYMBOL(sock_rfree);
1394 int sock_i_uid(struct sock *sk)
1398 read_lock_bh(&sk->sk_callback_lock);
1399 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1400 read_unlock_bh(&sk->sk_callback_lock);
1403 EXPORT_SYMBOL(sock_i_uid);
1405 unsigned long sock_i_ino(struct sock *sk)
1409 read_lock_bh(&sk->sk_callback_lock);
1410 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1411 read_unlock_bh(&sk->sk_callback_lock);
1414 EXPORT_SYMBOL(sock_i_ino);
1417 * Allocate a skb from the socket's send buffer.
1419 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1422 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1423 struct sk_buff *skb = alloc_skb(size, priority);
1425 skb_set_owner_w(skb, sk);
1431 EXPORT_SYMBOL(sock_wmalloc);
1434 * Allocate a skb from the socket's receive buffer.
1436 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1439 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1440 struct sk_buff *skb = alloc_skb(size, priority);
1442 skb_set_owner_r(skb, sk);
1450 * Allocate a memory block from the socket's option memory buffer.
1452 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1454 if ((unsigned)size <= sysctl_optmem_max &&
1455 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1457 /* First do the add, to avoid the race if kmalloc
1460 atomic_add(size, &sk->sk_omem_alloc);
1461 mem = kmalloc(size, priority);
1464 atomic_sub(size, &sk->sk_omem_alloc);
1468 EXPORT_SYMBOL(sock_kmalloc);
1471 * Free an option memory block.
1473 void sock_kfree_s(struct sock *sk, void *mem, int size)
1476 atomic_sub(size, &sk->sk_omem_alloc);
1478 EXPORT_SYMBOL(sock_kfree_s);
1480 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1481 I think, these locks should be removed for datagram sockets.
1483 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1487 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1491 if (signal_pending(current))
1493 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1494 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1495 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1497 if (sk->sk_shutdown & SEND_SHUTDOWN)
1501 timeo = schedule_timeout(timeo);
1503 finish_wait(sk_sleep(sk), &wait);
1509 * Generic send/receive buffer handlers
1512 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1513 unsigned long data_len, int noblock,
1516 struct sk_buff *skb;
1521 gfp_mask = sk->sk_allocation;
1522 if (gfp_mask & __GFP_WAIT)
1523 gfp_mask |= __GFP_REPEAT;
1525 timeo = sock_sndtimeo(sk, noblock);
1527 err = sock_error(sk);
1532 if (sk->sk_shutdown & SEND_SHUTDOWN)
1535 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1536 skb = alloc_skb(header_len, gfp_mask);
1541 /* No pages, we're done... */
1545 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1546 skb->truesize += data_len;
1547 skb_shinfo(skb)->nr_frags = npages;
1548 for (i = 0; i < npages; i++) {
1551 page = alloc_pages(sk->sk_allocation, 0);
1554 skb_shinfo(skb)->nr_frags = i;
1559 __skb_fill_page_desc(skb, i,
1561 (data_len >= PAGE_SIZE ?
1564 data_len -= PAGE_SIZE;
1567 /* Full success... */
1573 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1574 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1578 if (signal_pending(current))
1580 timeo = sock_wait_for_wmem(sk, timeo);
1583 skb_set_owner_w(skb, sk);
1587 err = sock_intr_errno(timeo);
1592 EXPORT_SYMBOL(sock_alloc_send_pskb);
1594 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1595 int noblock, int *errcode)
1597 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1599 EXPORT_SYMBOL(sock_alloc_send_skb);
1601 static void __lock_sock(struct sock *sk)
1602 __releases(&sk->sk_lock.slock)
1603 __acquires(&sk->sk_lock.slock)
1608 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1609 TASK_UNINTERRUPTIBLE);
1610 spin_unlock_bh(&sk->sk_lock.slock);
1612 spin_lock_bh(&sk->sk_lock.slock);
1613 if (!sock_owned_by_user(sk))
1616 finish_wait(&sk->sk_lock.wq, &wait);
1619 static void __release_sock(struct sock *sk)
1620 __releases(&sk->sk_lock.slock)
1621 __acquires(&sk->sk_lock.slock)
1623 struct sk_buff *skb = sk->sk_backlog.head;
1626 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1630 struct sk_buff *next = skb->next;
1632 WARN_ON_ONCE(skb_dst_is_noref(skb));
1634 sk_backlog_rcv(sk, skb);
1637 * We are in process context here with softirqs
1638 * disabled, use cond_resched_softirq() to preempt.
1639 * This is safe to do because we've taken the backlog
1642 cond_resched_softirq();
1645 } while (skb != NULL);
1648 } while ((skb = sk->sk_backlog.head) != NULL);
1651 * Doing the zeroing here guarantee we can not loop forever
1652 * while a wild producer attempts to flood us.
1654 sk->sk_backlog.len = 0;
1658 * sk_wait_data - wait for data to arrive at sk_receive_queue
1659 * @sk: sock to wait on
1660 * @timeo: for how long
1662 * Now socket state including sk->sk_err is changed only under lock,
1663 * hence we may omit checks after joining wait queue.
1664 * We check receive queue before schedule() only as optimization;
1665 * it is very likely that release_sock() added new data.
1667 int sk_wait_data(struct sock *sk, long *timeo)
1672 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1673 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1674 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1675 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1676 finish_wait(sk_sleep(sk), &wait);
1679 EXPORT_SYMBOL(sk_wait_data);
1682 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1684 * @size: memory size to allocate
1685 * @kind: allocation type
1687 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1688 * rmem allocation. This function assumes that protocols which have
1689 * memory_pressure use sk_wmem_queued as write buffer accounting.
1691 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1693 struct proto *prot = sk->sk_prot;
1694 int amt = sk_mem_pages(size);
1697 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1698 allocated = atomic_long_add_return(amt, prot->memory_allocated);
1701 if (allocated <= prot->sysctl_mem[0]) {
1702 if (prot->memory_pressure && *prot->memory_pressure)
1703 *prot->memory_pressure = 0;
1707 /* Under pressure. */
1708 if (allocated > prot->sysctl_mem[1])
1709 if (prot->enter_memory_pressure)
1710 prot->enter_memory_pressure(sk);
1712 /* Over hard limit. */
1713 if (allocated > prot->sysctl_mem[2])
1714 goto suppress_allocation;
1716 /* guarantee minimum buffer size under pressure */
1717 if (kind == SK_MEM_RECV) {
1718 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1720 } else { /* SK_MEM_SEND */
1721 if (sk->sk_type == SOCK_STREAM) {
1722 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1724 } else if (atomic_read(&sk->sk_wmem_alloc) <
1725 prot->sysctl_wmem[0])
1729 if (prot->memory_pressure) {
1732 if (!*prot->memory_pressure)
1734 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1735 if (prot->sysctl_mem[2] > alloc *
1736 sk_mem_pages(sk->sk_wmem_queued +
1737 atomic_read(&sk->sk_rmem_alloc) +
1738 sk->sk_forward_alloc))
1742 suppress_allocation:
1744 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1745 sk_stream_moderate_sndbuf(sk);
1747 /* Fail only if socket is _under_ its sndbuf.
1748 * In this case we cannot block, so that we have to fail.
1750 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1754 trace_sock_exceed_buf_limit(sk, prot, allocated);
1756 /* Alas. Undo changes. */
1757 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1758 atomic_long_sub(amt, prot->memory_allocated);
1761 EXPORT_SYMBOL(__sk_mem_schedule);
1764 * __sk_reclaim - reclaim memory_allocated
1767 void __sk_mem_reclaim(struct sock *sk)
1769 struct proto *prot = sk->sk_prot;
1771 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1772 prot->memory_allocated);
1773 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1775 if (prot->memory_pressure && *prot->memory_pressure &&
1776 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1777 *prot->memory_pressure = 0;
1779 EXPORT_SYMBOL(__sk_mem_reclaim);
1783 * Set of default routines for initialising struct proto_ops when
1784 * the protocol does not support a particular function. In certain
1785 * cases where it makes no sense for a protocol to have a "do nothing"
1786 * function, some default processing is provided.
1789 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1793 EXPORT_SYMBOL(sock_no_bind);
1795 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1800 EXPORT_SYMBOL(sock_no_connect);
1802 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1806 EXPORT_SYMBOL(sock_no_socketpair);
1808 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1812 EXPORT_SYMBOL(sock_no_accept);
1814 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1819 EXPORT_SYMBOL(sock_no_getname);
1821 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1825 EXPORT_SYMBOL(sock_no_poll);
1827 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1831 EXPORT_SYMBOL(sock_no_ioctl);
1833 int sock_no_listen(struct socket *sock, int backlog)
1837 EXPORT_SYMBOL(sock_no_listen);
1839 int sock_no_shutdown(struct socket *sock, int how)
1843 EXPORT_SYMBOL(sock_no_shutdown);
1845 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1846 char __user *optval, unsigned int optlen)
1850 EXPORT_SYMBOL(sock_no_setsockopt);
1852 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1853 char __user *optval, int __user *optlen)
1857 EXPORT_SYMBOL(sock_no_getsockopt);
1859 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1864 EXPORT_SYMBOL(sock_no_sendmsg);
1866 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1867 size_t len, int flags)
1871 EXPORT_SYMBOL(sock_no_recvmsg);
1873 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1875 /* Mirror missing mmap method error code */
1878 EXPORT_SYMBOL(sock_no_mmap);
1880 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1883 struct msghdr msg = {.msg_flags = flags};
1885 char *kaddr = kmap(page);
1886 iov.iov_base = kaddr + offset;
1888 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1892 EXPORT_SYMBOL(sock_no_sendpage);
1895 * Default Socket Callbacks
1898 static void sock_def_wakeup(struct sock *sk)
1900 struct socket_wq *wq;
1903 wq = rcu_dereference(sk->sk_wq);
1904 if (wq_has_sleeper(wq))
1905 wake_up_interruptible_all(&wq->wait);
1909 static void sock_def_error_report(struct sock *sk)
1911 struct socket_wq *wq;
1914 wq = rcu_dereference(sk->sk_wq);
1915 if (wq_has_sleeper(wq))
1916 wake_up_interruptible_poll(&wq->wait, POLLERR);
1917 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1921 static void sock_def_readable(struct sock *sk, int len)
1923 struct socket_wq *wq;
1926 wq = rcu_dereference(sk->sk_wq);
1927 if (wq_has_sleeper(wq))
1928 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1929 POLLRDNORM | POLLRDBAND);
1930 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1934 static void sock_def_write_space(struct sock *sk)
1936 struct socket_wq *wq;
1940 /* Do not wake up a writer until he can make "significant"
1943 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1944 wq = rcu_dereference(sk->sk_wq);
1945 if (wq_has_sleeper(wq))
1946 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1947 POLLWRNORM | POLLWRBAND);
1949 /* Should agree with poll, otherwise some programs break */
1950 if (sock_writeable(sk))
1951 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1957 static void sock_def_destruct(struct sock *sk)
1959 kfree(sk->sk_protinfo);
1962 void sk_send_sigurg(struct sock *sk)
1964 if (sk->sk_socket && sk->sk_socket->file)
1965 if (send_sigurg(&sk->sk_socket->file->f_owner))
1966 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1968 EXPORT_SYMBOL(sk_send_sigurg);
1970 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1971 unsigned long expires)
1973 if (!mod_timer(timer, expires))
1976 EXPORT_SYMBOL(sk_reset_timer);
1978 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1980 if (timer_pending(timer) && del_timer(timer))
1983 EXPORT_SYMBOL(sk_stop_timer);
1985 void sock_init_data(struct socket *sock, struct sock *sk)
1987 skb_queue_head_init(&sk->sk_receive_queue);
1988 skb_queue_head_init(&sk->sk_write_queue);
1989 skb_queue_head_init(&sk->sk_error_queue);
1990 #ifdef CONFIG_NET_DMA
1991 skb_queue_head_init(&sk->sk_async_wait_queue);
1994 sk->sk_send_head = NULL;
1996 init_timer(&sk->sk_timer);
1998 sk->sk_allocation = GFP_KERNEL;
1999 sk->sk_rcvbuf = sysctl_rmem_default;
2000 sk->sk_sndbuf = sysctl_wmem_default;
2001 sk->sk_state = TCP_CLOSE;
2002 sk_set_socket(sk, sock);
2004 sock_set_flag(sk, SOCK_ZAPPED);
2007 sk->sk_type = sock->type;
2008 sk->sk_wq = sock->wq;
2013 spin_lock_init(&sk->sk_dst_lock);
2014 rwlock_init(&sk->sk_callback_lock);
2015 lockdep_set_class_and_name(&sk->sk_callback_lock,
2016 af_callback_keys + sk->sk_family,
2017 af_family_clock_key_strings[sk->sk_family]);
2019 sk->sk_state_change = sock_def_wakeup;
2020 sk->sk_data_ready = sock_def_readable;
2021 sk->sk_write_space = sock_def_write_space;
2022 sk->sk_error_report = sock_def_error_report;
2023 sk->sk_destruct = sock_def_destruct;
2025 sk->sk_sndmsg_page = NULL;
2026 sk->sk_sndmsg_off = 0;
2028 sk->sk_peer_pid = NULL;
2029 sk->sk_peer_cred = NULL;
2030 sk->sk_write_pending = 0;
2031 sk->sk_rcvlowat = 1;
2032 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2033 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2035 sk->sk_stamp = ktime_set(-1L, 0);
2038 * Before updating sk_refcnt, we must commit prior changes to memory
2039 * (Documentation/RCU/rculist_nulls.txt for details)
2042 atomic_set(&sk->sk_refcnt, 1);
2043 atomic_set(&sk->sk_drops, 0);
2045 EXPORT_SYMBOL(sock_init_data);
2047 void lock_sock_nested(struct sock *sk, int subclass)
2050 spin_lock_bh(&sk->sk_lock.slock);
2051 if (sk->sk_lock.owned)
2053 sk->sk_lock.owned = 1;
2054 spin_unlock(&sk->sk_lock.slock);
2056 * The sk_lock has mutex_lock() semantics here:
2058 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2061 EXPORT_SYMBOL(lock_sock_nested);
2063 void release_sock(struct sock *sk)
2066 * The sk_lock has mutex_unlock() semantics:
2068 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2070 spin_lock_bh(&sk->sk_lock.slock);
2071 if (sk->sk_backlog.tail)
2073 sk->sk_lock.owned = 0;
2074 if (waitqueue_active(&sk->sk_lock.wq))
2075 wake_up(&sk->sk_lock.wq);
2076 spin_unlock_bh(&sk->sk_lock.slock);
2078 EXPORT_SYMBOL(release_sock);
2081 * lock_sock_fast - fast version of lock_sock
2084 * This version should be used for very small section, where process wont block
2085 * return false if fast path is taken
2086 * sk_lock.slock locked, owned = 0, BH disabled
2087 * return true if slow path is taken
2088 * sk_lock.slock unlocked, owned = 1, BH enabled
2090 bool lock_sock_fast(struct sock *sk)
2093 spin_lock_bh(&sk->sk_lock.slock);
2095 if (!sk->sk_lock.owned)
2097 * Note : We must disable BH
2102 sk->sk_lock.owned = 1;
2103 spin_unlock(&sk->sk_lock.slock);
2105 * The sk_lock has mutex_lock() semantics here:
2107 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2111 EXPORT_SYMBOL(lock_sock_fast);
2113 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2116 if (!sock_flag(sk, SOCK_TIMESTAMP))
2117 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2118 tv = ktime_to_timeval(sk->sk_stamp);
2119 if (tv.tv_sec == -1)
2121 if (tv.tv_sec == 0) {
2122 sk->sk_stamp = ktime_get_real();
2123 tv = ktime_to_timeval(sk->sk_stamp);
2125 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2127 EXPORT_SYMBOL(sock_get_timestamp);
2129 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2132 if (!sock_flag(sk, SOCK_TIMESTAMP))
2133 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2134 ts = ktime_to_timespec(sk->sk_stamp);
2135 if (ts.tv_sec == -1)
2137 if (ts.tv_sec == 0) {
2138 sk->sk_stamp = ktime_get_real();
2139 ts = ktime_to_timespec(sk->sk_stamp);
2141 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2143 EXPORT_SYMBOL(sock_get_timestampns);
2145 void sock_enable_timestamp(struct sock *sk, int flag)
2147 if (!sock_flag(sk, flag)) {
2148 sock_set_flag(sk, flag);
2150 * we just set one of the two flags which require net
2151 * time stamping, but time stamping might have been on
2152 * already because of the other one
2155 flag == SOCK_TIMESTAMP ?
2156 SOCK_TIMESTAMPING_RX_SOFTWARE :
2158 net_enable_timestamp();
2163 * Get a socket option on an socket.
2165 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2166 * asynchronous errors should be reported by getsockopt. We assume
2167 * this means if you specify SO_ERROR (otherwise whats the point of it).
2169 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2170 char __user *optval, int __user *optlen)
2172 struct sock *sk = sock->sk;
2174 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2176 EXPORT_SYMBOL(sock_common_getsockopt);
2178 #ifdef CONFIG_COMPAT
2179 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2180 char __user *optval, int __user *optlen)
2182 struct sock *sk = sock->sk;
2184 if (sk->sk_prot->compat_getsockopt != NULL)
2185 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2187 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2189 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2192 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2193 struct msghdr *msg, size_t size, int flags)
2195 struct sock *sk = sock->sk;
2199 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2200 flags & ~MSG_DONTWAIT, &addr_len);
2202 msg->msg_namelen = addr_len;
2205 EXPORT_SYMBOL(sock_common_recvmsg);
2208 * Set socket options on an inet socket.
2210 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2211 char __user *optval, unsigned int optlen)
2213 struct sock *sk = sock->sk;
2215 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2217 EXPORT_SYMBOL(sock_common_setsockopt);
2219 #ifdef CONFIG_COMPAT
2220 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2221 char __user *optval, unsigned int optlen)
2223 struct sock *sk = sock->sk;
2225 if (sk->sk_prot->compat_setsockopt != NULL)
2226 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2228 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2230 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2233 void sk_common_release(struct sock *sk)
2235 if (sk->sk_prot->destroy)
2236 sk->sk_prot->destroy(sk);
2239 * Observation: when sock_common_release is called, processes have
2240 * no access to socket. But net still has.
2241 * Step one, detach it from networking:
2243 * A. Remove from hash tables.
2246 sk->sk_prot->unhash(sk);
2249 * In this point socket cannot receive new packets, but it is possible
2250 * that some packets are in flight because some CPU runs receiver and
2251 * did hash table lookup before we unhashed socket. They will achieve
2252 * receive queue and will be purged by socket destructor.
2254 * Also we still have packets pending on receive queue and probably,
2255 * our own packets waiting in device queues. sock_destroy will drain
2256 * receive queue, but transmitted packets will delay socket destruction
2257 * until the last reference will be released.
2262 xfrm_sk_free_policy(sk);
2264 sk_refcnt_debug_release(sk);
2267 EXPORT_SYMBOL(sk_common_release);
2269 static DEFINE_RWLOCK(proto_list_lock);
2270 static LIST_HEAD(proto_list);
2272 #ifdef CONFIG_PROC_FS
2273 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2275 int val[PROTO_INUSE_NR];
2278 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2280 #ifdef CONFIG_NET_NS
2281 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2283 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2285 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2287 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2289 int cpu, idx = prot->inuse_idx;
2292 for_each_possible_cpu(cpu)
2293 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2295 return res >= 0 ? res : 0;
2297 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2299 static int __net_init sock_inuse_init_net(struct net *net)
2301 net->core.inuse = alloc_percpu(struct prot_inuse);
2302 return net->core.inuse ? 0 : -ENOMEM;
2305 static void __net_exit sock_inuse_exit_net(struct net *net)
2307 free_percpu(net->core.inuse);
2310 static struct pernet_operations net_inuse_ops = {
2311 .init = sock_inuse_init_net,
2312 .exit = sock_inuse_exit_net,
2315 static __init int net_inuse_init(void)
2317 if (register_pernet_subsys(&net_inuse_ops))
2318 panic("Cannot initialize net inuse counters");
2323 core_initcall(net_inuse_init);
2325 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2327 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2329 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2331 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2333 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2335 int cpu, idx = prot->inuse_idx;
2338 for_each_possible_cpu(cpu)
2339 res += per_cpu(prot_inuse, cpu).val[idx];
2341 return res >= 0 ? res : 0;
2343 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2346 static void assign_proto_idx(struct proto *prot)
2348 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2350 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2351 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2355 set_bit(prot->inuse_idx, proto_inuse_idx);
2358 static void release_proto_idx(struct proto *prot)
2360 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2361 clear_bit(prot->inuse_idx, proto_inuse_idx);
2364 static inline void assign_proto_idx(struct proto *prot)
2368 static inline void release_proto_idx(struct proto *prot)
2373 int proto_register(struct proto *prot, int alloc_slab)
2376 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2377 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2380 if (prot->slab == NULL) {
2381 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2386 if (prot->rsk_prot != NULL) {
2387 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2388 if (prot->rsk_prot->slab_name == NULL)
2389 goto out_free_sock_slab;
2391 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2392 prot->rsk_prot->obj_size, 0,
2393 SLAB_HWCACHE_ALIGN, NULL);
2395 if (prot->rsk_prot->slab == NULL) {
2396 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2398 goto out_free_request_sock_slab_name;
2402 if (prot->twsk_prot != NULL) {
2403 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2405 if (prot->twsk_prot->twsk_slab_name == NULL)
2406 goto out_free_request_sock_slab;
2408 prot->twsk_prot->twsk_slab =
2409 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2410 prot->twsk_prot->twsk_obj_size,
2412 SLAB_HWCACHE_ALIGN |
2415 if (prot->twsk_prot->twsk_slab == NULL)
2416 goto out_free_timewait_sock_slab_name;
2420 write_lock(&proto_list_lock);
2421 list_add(&prot->node, &proto_list);
2422 assign_proto_idx(prot);
2423 write_unlock(&proto_list_lock);
2426 out_free_timewait_sock_slab_name:
2427 kfree(prot->twsk_prot->twsk_slab_name);
2428 out_free_request_sock_slab:
2429 if (prot->rsk_prot && prot->rsk_prot->slab) {
2430 kmem_cache_destroy(prot->rsk_prot->slab);
2431 prot->rsk_prot->slab = NULL;
2433 out_free_request_sock_slab_name:
2435 kfree(prot->rsk_prot->slab_name);
2437 kmem_cache_destroy(prot->slab);
2442 EXPORT_SYMBOL(proto_register);
2444 void proto_unregister(struct proto *prot)
2446 write_lock(&proto_list_lock);
2447 release_proto_idx(prot);
2448 list_del(&prot->node);
2449 write_unlock(&proto_list_lock);
2451 if (prot->slab != NULL) {
2452 kmem_cache_destroy(prot->slab);
2456 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2457 kmem_cache_destroy(prot->rsk_prot->slab);
2458 kfree(prot->rsk_prot->slab_name);
2459 prot->rsk_prot->slab = NULL;
2462 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2463 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2464 kfree(prot->twsk_prot->twsk_slab_name);
2465 prot->twsk_prot->twsk_slab = NULL;
2468 EXPORT_SYMBOL(proto_unregister);
2470 #ifdef CONFIG_PROC_FS
2471 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2472 __acquires(proto_list_lock)
2474 read_lock(&proto_list_lock);
2475 return seq_list_start_head(&proto_list, *pos);
2478 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2480 return seq_list_next(v, &proto_list, pos);
2483 static void proto_seq_stop(struct seq_file *seq, void *v)
2484 __releases(proto_list_lock)
2486 read_unlock(&proto_list_lock);
2489 static char proto_method_implemented(const void *method)
2491 return method == NULL ? 'n' : 'y';
2494 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2496 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2497 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2500 sock_prot_inuse_get(seq_file_net(seq), proto),
2501 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2502 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2504 proto->slab == NULL ? "no" : "yes",
2505 module_name(proto->owner),
2506 proto_method_implemented(proto->close),
2507 proto_method_implemented(proto->connect),
2508 proto_method_implemented(proto->disconnect),
2509 proto_method_implemented(proto->accept),
2510 proto_method_implemented(proto->ioctl),
2511 proto_method_implemented(proto->init),
2512 proto_method_implemented(proto->destroy),
2513 proto_method_implemented(proto->shutdown),
2514 proto_method_implemented(proto->setsockopt),
2515 proto_method_implemented(proto->getsockopt),
2516 proto_method_implemented(proto->sendmsg),
2517 proto_method_implemented(proto->recvmsg),
2518 proto_method_implemented(proto->sendpage),
2519 proto_method_implemented(proto->bind),
2520 proto_method_implemented(proto->backlog_rcv),
2521 proto_method_implemented(proto->hash),
2522 proto_method_implemented(proto->unhash),
2523 proto_method_implemented(proto->get_port),
2524 proto_method_implemented(proto->enter_memory_pressure));
2527 static int proto_seq_show(struct seq_file *seq, void *v)
2529 if (v == &proto_list)
2530 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2539 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2541 proto_seq_printf(seq, list_entry(v, struct proto, node));
2545 static const struct seq_operations proto_seq_ops = {
2546 .start = proto_seq_start,
2547 .next = proto_seq_next,
2548 .stop = proto_seq_stop,
2549 .show = proto_seq_show,
2552 static int proto_seq_open(struct inode *inode, struct file *file)
2554 return seq_open_net(inode, file, &proto_seq_ops,
2555 sizeof(struct seq_net_private));
2558 static const struct file_operations proto_seq_fops = {
2559 .owner = THIS_MODULE,
2560 .open = proto_seq_open,
2562 .llseek = seq_lseek,
2563 .release = seq_release_net,
2566 static __net_init int proto_init_net(struct net *net)
2568 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2574 static __net_exit void proto_exit_net(struct net *net)
2576 proc_net_remove(net, "protocols");
2580 static __net_initdata struct pernet_operations proto_net_ops = {
2581 .init = proto_init_net,
2582 .exit = proto_exit_net,
2585 static int __init proto_init(void)
2587 return register_pernet_subsys(&proto_net_ops);
2590 subsys_initcall(proto_init);
2592 #endif /* PROC_FS */